Extract features from FPN neural network ,then visualize features by inverse quantifying and turning those feature tensors into PNG files.
This project is mainly used for extracting features from FPN structure in detectron2 , maybe applying it to JDE and yolo v3 framework someday.
This work will split into two parts , first one is to dump the feature tensors , second one is to turn these feature tensors into visualized PNG files.
Remember to follow the instruct written in the start.md